Exploring influencing factors of health resilience for urban buildings by integrated CHATGPT-empowered BERTopic model: A case study of Hong Kong

Abstract

Enhancing building health resilience (BHR) is a crucial pathway to mitigate people's health loss under natural or manmade disturbances. However, as BHR is quite a new concept, previous research lacks a comprehensive investigation and deep understanding of BHR influencing factors. Topic modeling method is innovative to extract topics from multi-source data, including literature, news, reports and other unstructured online data, which could fill the gap of lacking sufficient literatures and other sources support. This study aims to explore BHR influencing factors by integrating and literature review-based identification and topic modeling method. Due to ChatGPT's exceptional ability to extract information from unstructured text data, an integrated ChatGPT-empowered BERTopic (BERTGPT) model is proposed for multi-source exploration, exploring BHR influencing factors by twice ChatGPT empowerment in BERTopic, which can act as a supplementary of literature-based identification. Results show that BHR influencing factors comes from four dimensions: building attributes, building environment, building demographics, and human behavior. Furthermore, this model was validated by classification accuracy and summarization precision, demonstrating the model's effectiveness in extracting representative topics from multi-source unstructured data. This study integrated the factors identified from the literature and multi-source data, providing a clear direction for BHR enhancement. This study also develops a novel AI-enabled approach for exploring potential factors influencing BHR and other emerging concepts lacking sufficient literature support, utilizing multi-source unstructured data.